#!/usr/bin/env python """ AppServer The WebKit app server is a TCP/IP server that accepts requests, hands them off to the Application and sends the request back over the connection. The fact that the app server stays resident is what makes it so much quicker than traditional CGI programming. Everything gets cached. FUTURE * Implement the additional settings that are commented out below. """ from Common import * from AutoReloadingAppServer import AutoReloadingAppServer as AppServer from MiscUtils.Funcs import timestamp from marshal import dumps, loads import os, sys from threading import Lock, Thread, Event import threading import Queue import select import socket import threading import time import errno import traceback from WebUtils import Funcs debug = 0 DefaultConfig = { 'Port': 8086, 'MaxServerThreads': 20, 'MinServerThreads': 5, 'StartServerThreads': 10, # @@ 2000-04-27 ce: None of the following settings are implemented # 'RequestQueueSize': 16,# 'RequestBufferSize': 64*1024, # 'SocketType': 'inet', # inet, unix } #Below used with the RestartApplication function #ReStartLock=Lock() #ReqCount=0 #Need to know this value for communications #Note that this limits the size of the dictionary we receive from the AppServer to 2,147,483,647 bytes int_length = len(dumps(int(1))) server = None class ThreadedAppServer(AppServer): """ """ ## Init ## def __init__(self, path=None): AppServer.__init__(self, path) self._addr = None threadCount = self.setting('StartServerThreads') self.maxServerThreads = self.setting('MaxServerThreads') self.minServerThreads = self.setting('MinServerThreads') self.monitorPort = None self.threadPool = [] self.threadCount=0 self.threadUseCounter=[] self.requestQueue = Queue.Queue(self.maxServerThreads * 2) # twice the number of threads we have self.rhCache = [] # This will grow to a limit of the number of # threads plus the size of the requestQueue plus one. # It used to be a Queue but since we don't make # use of the blocking behavior and because of problems # with Queue.Empty being raised on .get_nowait() # when the queue isn't in fact empty, we have switched # to using a list instead. self.mainsocket = socket.socket(socket.AF_INET, socket.SOCK_STREAM) # Must use SO_REUSEADDR to avoid problems restarting the app server # This was discussed on Webware-devel in Oct 2001, and this solution # was found by Jeff Johnson self.mainsocket.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1) addr = self.address() try: self.mainsocket.bind(addr) except: if self.running: self.initiateShutdown() self._closeThread.join() raise print "Listening on", addr open(self.serverSidePath('address.text'), 'w').write('%s:%d' % (addr[0], addr[1])) self.monitorPort = addr[1]-1 out = sys.stdout out.write('Creating %d threads' % threadCount) for i in range(threadCount): self.spawnThread() out.write(".") out.flush() out.write("\n") # @@ 2001-05-30 ce: another hard coded number: @@jsl- not everything needs to be configurable.... self.mainsocket.listen(1024) self.recordPID() self.readyForRequests() def isPersistent(self): return 1 def mainloop(self, monitor=None, timeout=1): from errno import EINTR inputsockets = [self.mainsocket,] if monitor: inputsockets.append(monitor.insock) threadCheckInterval = self.maxServerThreads*2 threadUpdateDivisor = 5 #grabstat interval threadCheck=0 while 1: if not self.running: return #block for timeout seconds waiting for connections try: input, output, exc = select.select(inputsockets,[],[],timeout) except select.error, v: # if the error is EINTR/interrupt, then self.running should be set to 0 and # we'll exit on the next loop if v[0] == EINTR or v[0]==0: break else: raise # @@ gat 2002-03-20: I found through heavy testing of restart behavior that WebKit # dropped fewer requests on restart by removing this test. ## if not self.running: ## return for sock in input: if sock.getsockname()[1] == self.monitorPort: client,addr = sock.accept() monitor.activate(client) self.requestQueue.put(monitor) else: self._reqCount = self._reqCount+1 rh = None client,addr = sock.accept() try: rh = self.rhCache.pop() except IndexError: rh = RequestHandler(self) rh.activate(client, self._reqCount) self.requestQueue.put(rh) if threadCheck % threadUpdateDivisor == 0: self.updateThreadUsage() if threadCheck > threadCheckInterval: if debug: print "\nBusy Threads: ", self.activeThreadCount() threadCheck=0 self.manageThreadCount() else: threadCheck = threadCheck+1 self.restartIfNecessary() def activeThreadCount(self): """ Get a snapshot of the number of threads currently in use. """ count=0 for i in self.threadPool: if i.processing: count = count+1 return count def updateThreadUsage(self): """ Update the threadUseCounter list. """ count = self.activeThreadCount() if len(self.threadUseCounter) > self.maxServerThreads: self.threadUseCounter.pop(0) self.threadUseCounter.append(count) def manageThreadCount(self): """ Adjust the number of threads in use. This algorithm needs work. The edges (ie at the minserverthreads) are tricky. When working with this, remember thread creation is CHEAP """ avg=0 max=0 debug=0 if debug: print "ThreadUse Samples=%s" % str(self.threadUseCounter) for i in self.threadUseCounter: avg = avg + i if i > max: max = i avg = avg / len(self.threadUseCounter) if debug: print "Average Thread Use: ", avg if debug: print "Max Thread Use: ", max if debug: print "ThreadCount: ", self.threadCount if len(self.threadUseCounter) < self.maxServerThreads: return #not enough samples margin = self.threadCount / 2 #smoothing factor if debug: print "margin=", margin if avg > self.threadCount - margin and self.threadCount < self.maxServerThreads: # Running low: double thread count n = min(self.threadCount, self.maxServerThreads-self.threadCount) if debug: print "Adding %s threads" % n for i in range(n): self.spawnThread() elif avg < self.threadCount - margin and self.threadCount > self.minServerThreads: n=min(self.threadCount - self.minServerThreads, self.threadCount - max) self.absorbThread(n) else: #cleanup any stale threads that we killed but haven't joined self.absorbThread(0) def spawnThread(self): debug=0 if debug: print "Spawning new thread" t = Thread(target=self.threadloop) t.processing=0 t.start() self.threadPool.append(t) self.threadCount = self.threadCount+1 if debug: print "New Thread Spawned, threadCount=", self.threadCount ## self.threadUseCounter=[] #reset def absorbThread(self,count=1): """ Absorb a thread. We do this by putting a None on the Queue. When a thread gets it, that tells it to exit. BUT, even though we put it on, the thread may not have retrieved it before we exit this function. So we need to decrement the thread count even if we didn't find a thread that isn't alive. We'll get it the next time through. """ debug=0 if debug: print "Absorbing %s Threads" % count for i in range(count): self.requestQueue.put(None) self.threadCount = self.threadCount-1 for i in self.threadPool: if not i.isAlive(): rv=i.join() #Don't need a timeout, it isn't alive self.threadPool.remove(i) if debug: print "Thread Absorbed, Real Thread Count=", len(self.threadPool) ## self.threadUseCounter=[] #reset def threadloop(self): self.initThread() t=threading.currentThread() t.processing=0 try: while 1: try: rh=self.requestQueue.get() if rh == None: #None means time to quit if debug: print "Thread retrieved None, quitting" break # @@ gat 2002-03-21: we don't want to drop a request that we've # already received from the adapter; therefore, I removed the # test of self.running. Now, once a request gets put into the # request queue, it WILL be processed, never dropped. This # might slow down shutdown a bit if many requests are currently # being processed, but it makes restarting the appserver work # better. ## if self.running: ## t.processing=1 ## try: ## rh.handleRequest() ## except: ## traceback.print_exc(file=sys.stderr) ## t.processing=0 t.processing=1 try: rh.handleRequest() except: traceback.print_exc(file=sys.stderr) t.processing=0 rh.close() except Queue.Empty: pass finally: self.delThread() if debug: print threading.currentThread(), "Quitting" def initThread(self): """ Invoked immediately by threadloop() as a hook for subclasses. This implementation does nothing and subclasses need not invoke super. """ pass def delThread(self): """ Invoked immediately by threadloop() as a hook for subclasses. This implementation does nothing and subclasses need not invoke super. """ pass def awakeSelect(self): """ Send a connect to ourself to pop the select() call out of it's loop safely """ sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM) addr = self.address() try: sock.connect(addr) sock.close() except: pass return def shutDown(self): self.running=0 self.awakeSelect() self._shuttingdown=1 #jsl-is this used anywhere? print "ThreadedAppServer: Shutting Down" self.mainsocket.close() for i in range(self.threadCount): self.requestQueue.put(None)#kill all threads for i in self.threadPool: try: i.join() except: pass AppServer.shutDown(self) ## Network Server ## def address(self): if self._addr is None: self._addr = (self.setting('Host'), self.setting('Port')) return self._addr ## Misc ## def setRequestQueueSize(self, value): assert value>=1 self.__class__.request_queue_size = value class Monitor: def __init__(self, server): self.server = server self.port = server.monitorPort self.insock = socket.socket(socket.AF_INET, socket.SOCK_STREAM) self.insock.bind((server.address()[0],server.address()[1]-1)) self.insock.listen(1) print "******** Listening to Monitor Socket ************" def activate(self, socket): self.sock = socket def close(self): self.sock = None def handleRequest(self): verbose = self.server._verbose startTime = time.time() if verbose: print 'BEGIN REQUEST' print time.asctime(time.localtime(startTime)) conn = self.sock if verbose: print 'receiving request from', conn BUFSIZE = 8*1024 chunk = '' missing = int_length while missing > 0: block = conn.recv(missing) if not block: conn.close() raise NotEnoughDataError, 'received only %d out of %d bytes when receiving dict_length' % (len(chunk), int_length) chunk = chunk + block missing = int_length - len(chunk) dict_length = loads(chunk) if type(dict_length) != type(1): conn.close() print "Error: Invalid AppServer protocol" return 0 chunk = '' missing = dict_length while missing > 0: block = conn.recv(missing) if not block: conn.close() raise NotEnoughDataError, 'received only %d out of %d bytes when receiving dict' % (len(chunk), dict_length) chunk = chunk + block missing = dict_length - len(chunk) dict = loads(chunk) if dict['format'] == "STATUS": conn.send(str(self.server._reqCount)) if dict['format'] == 'QUIT': conn.send("OK") conn.close() self.server.shutDown() from WebKit.ASStreamOut import ASStreamOut class TASASStreamOut(ASStreamOut): def __init__(self, sock): ASStreamOut.__init__(self) self._socket = sock def flush(self): debug=0 result = ASStreamOut.flush(self) if result: ##a true return value means we can send reslen = len(self._buffer) if debug: print "TASASStreamout is sending %s bytes" % reslen sent = 0 while sent < reslen: try: sent = sent + self._socket.send(self._buffer[sent:sent+8192]) except socket.error, e: if e[0]==errno.EPIPE: #broken pipe pass else: print "StreamOut Error: ", e break self.pop(sent) class RequestHandler: def __init__(self, server): self.server = server def activate(self, sock, number): """ Activates the handler for processing the request. Number is the number of the request, mostly used to identify verbose output. Each request should be given a unique, incremental number. """ self.sock = sock # self._strmOut = TASASStreamOut(sock) self._number = number def close(self): self.sock = None # self._strmOut = None self.server.rhCache.append(self) def handleRequest(self): verbose = self.server._verbose startTime = time.time() if verbose: print '%5i %s ' % (self._number, timestamp()['pretty']), conn = self.sock # @@ 2001-05-30 ce: Ack! Look at this hard coding. BUFSIZE = 8*1024 data = [] chunk = '' missing = int_length while missing > 0: block = conn.recv(missing) if not block: conn.close() if len(chunk) == 0: # We probably awakened due to awakeSelect being called. return 0 else: # We got a partial request -- something went wrong. raise NotEnoughDataError, 'received only %d out of %d bytes when receiving dict_length' % (len(chunk), int_length) chunk = chunk + block missing = int_length - len(chunk) dict_length = loads(chunk) if type(dict_length) != type(1): conn.close() print print "Error: Invalid AppServer protocol" return 0 chunk = '' missing = dict_length while missing > 0: block = conn.recv(missing) if not block: conn.close() raise NotEnoughDataError, 'received only %d out of %d bytes when receiving dict' % (len(chunk), dict_length) chunk = chunk + block missing = dict_length - len(chunk) dict = loads(chunk) #if verbose: print "Comm Delay=%s" % (time.time() - dict['time']) if dict: if verbose: if dict.has_key('environ'): requestURI = Funcs.requestURI(dict['environ']) else: requestURI = None print requestURI dict['input'] = conn.makefile("rb",8012) strmOut = TASASStreamOut(self.sock) transaction = self.server._app.dispatchRawRequest(dict, strmOut) strmOut.close() try: conn.shutdown(1) conn.close() except: pass if verbose: duration = '%0.2f secs' % (time.time() - startTime) duration = string.ljust(duration, 19) print '%5i %s %s' % (self._number, duration, requestURI) print transaction._application=None transaction.die() del transaction def restartApp(self): """ Not used """ if self.server.num_requests> 200: print "Trying to get lock" ReStartLock.acquire() if self.server.num_requests> 200: #check again to make sure another thread didn't do it print "Restarting Application" currApp=self.server.wkApp wkAppServer=currApp._server newApp = wkAppServer.createApplication() newApp._sessions = currApp._sessions wkAppServer._app=newApp self.server.wkApp=newApp for i in currApp._factoryList: currApp._factoryList.remove(i) for i in currApp._factoryByExt.keys(): currApp._factoryByExt[i]=None currApp._canFactory=None wkAppServer._plugIns=[] wkAppServer.loadPlugIns() self.server.num_requests=0 print "Refs to old App=",sys.getrefcount(currApp) currApp=None ReStartLock.release() # This will be thrown when less data arrived on the socket than we were expecting. class NotEnoughDataError(Exception): pass def run(useMonitor = 0, workDir=None): global server global monitor monitor = useMonitor try: server = None server = ThreadedAppServer(workDir) if useMonitor: monitor_socket = Monitor(server) else: monitor_socket = None # On NT, run mainloop in a different thread because it's not safe for # Ctrl-C to be caught while manipulating the queues. # It's not safe on Linux either, but there, it appears that Ctrl-C # will trigger an exception in ANY thread, so this fix doesn't help. if os.name == 'nt': # catch the exception raised by sys.exit so that we can re-call it # in the main thread. global exitStatus exitStatus = None def windowsmainloop(server, monitor): global exitStatus try: server.mainloop(monitor) except SystemExit, e: exitStatus = e.code # Run the server thread t = threading.Thread(target=windowsmainloop, args=(server, monitor_socket)) t.start() try: while server.running: time.sleep(1.0) except KeyboardInterrupt: pass server.running = 0 t.join() # re-call sys.exit if necessary if exitStatus: sys.exit(exitStatus) else: try: server.mainloop(monitor_socket) except KeyboardInterrupt, e: server.shutDown() except Exception, e: if not isinstance(e, SystemExit): import traceback traceback.print_exc(file=sys.stderr) #print e print print "Exiting AppServer" if server: if server.running: server.initiateShutdown() server._closeThread.join() # if we're here as a result of exit() being called, # exit with that return code. if isinstance(e,SystemExit): sys.exit(e) sys.exit() def shutDown(arg1,arg2): global server print "Shutdown Called", time.asctime(time.localtime(time.time())) if server: server.initiateShutdown() else: print 'WARNING: No server reference to shutdown.' import signal signal.signal(signal.SIGINT, shutDown) signal.signal(signal.SIGTERM, shutDown) usage = """ The AppServer is the main process of WebKit. It handles requests for servlets from webservers. ThreadedAppServer takes the following command line arguments: stop: Stop the currently running Apperver. daemon: run as a daemon If AppServer is called with no arguments, it will start the AppServer and record the pid of the process in appserverpid.txt """ import re settingRE = re.compile(r'^--([a-zA-Z][a-zA-Z0-9]*\.[a-zA-Z][a-zA-Z0-9]*)=') from MiscUtils import Configurable def main(args): monitor=0 function=run daemon=0 workDir=None for i in args[:]: if settingRE.match(i): match = settingRE.match(i) name = match.group(1) value = i[match.end():] Configurable.addCommandLineSetting(name, value) elif i == "monitor": print "Enabling Monitoring" monitor=1 elif i == "stop": import AppServer function=AppServer.stop elif i == "daemon": daemon=1 elif i == "start": pass elif i[:8] == "workdir=": workDir = i[8:] else: print usage if daemon: if os.name == "posix": pid=os.fork() if pid: sys.exit() else: print "daemon mode not available on your OS" function(monitor, workDir)